Majority Party Bias in U.S. Congressional Conference Committees

Abstract

This article examines the representativeness of conference committees in the U.S. Congress by measuring the difference in observed policy preferences between the conference delegations and the parent bodies. We predict and find significant differences between the House and Senate in terms of the partisan bias of conference delegations. House conference delegations are systematically biased in favor of the majority party and away from the chamber median. We take the additional step of exploring the source of this bias. In particular, we examine whether majority party bias in conference is a function of partisan processes at work directly in the selection of conferees. We find evidence that the conditions of majority party influence in the House are consistent with some existing theoretical models of party influence in legislating. There is less conclusive evidence of partisan processes in the Senate, which is consistent with institutional differences in appointment practices.

Notes

1. It is worth noting that party ratios on standing committees, with the important exceptions of the House Committee on Rules and the two ethics committees, have closely tracked party ratios on the floor throughout most of the twentieth century (Poole and Rosenthal 1997). However, the majority party uses rounding (a fraction of a seat cannot be assigned) a bit more to its advantage. In the 111th Congress (2009–2010), for example, the Democrats had 59.1% of House seats and had 61.7, 61.5, 61.0, and 63.4% of the seats on Appropriations, Budget, Energy and Commerce, and Ways and Means, respectively.

2. While we commend Krehbiel (1991) for taking the first steps in examining this important question, his analysis considers less than half of the single-committee conferences in the House of Representatives in one Congress (the 99th Congress, 1985–1986). He excluded Appropriations, Energy and Commerce, and Ways and Means conferences that might be of keen interest to the parties. In addition, he gives no attention to the Senate. Furthermore, Krehbiel does not directly test the proposition that conferees are selected to move the medians of the conference delegations toward the policy location of most majority party members. This seems to us to be the operative question, and certainly the more relevant to Krehbiel's theory.

3. It is important to note that the conditional party government thesis was introduced in the context of the House (Rohde 1991), and it has been shown elsewhere that the argument applies less well in the Senate (Smith 2007).

4. We acknowledge that the periods of institutional change identified in the Party Eras Proposition closely correspond to important changes in leadership style. The Speakerships of O’Neill (1977–1986) and Gingrich (1995–1998) began periods of increasingly public and partisan Speakers (Harris 1998; Peters 1999). While it has been argued elsewhere that party leaders are agents of their parties, and that changes in leadership behavior can be explained by contextual factors (Cooper and Brady 1981; Sinclair 1999), it is plausible that differences in individual approaches to leadership may contribute to variation in conference appointment practices. Such an analysis would require theoretical attention beyond the scope of this article, but this remains a subject worthy of future examination.

5. We account for 87% of the population of conferences over the period of analysis. Using the periods outlined by the theory, we account for 90% of conferences in the first period (88th–93d Congress), 85% in the second period (94th–103d Congress), and 80% in the third period (104th–107th Congress).

6. Common space scores assume that legislators who serve in both houses have identical policy positions in both houses. This assumption anchors the estimations of policy positions and allows representatives and senators to be placed on the same scale.

7. It is important to note that even when the decision space is characterized by higher order dimensions, we would still expect the parties to be most active in influencing policy along the dimension most relevant to them (the first dimension). Even so, we undertook the additional exercise of examining the subset of conference committees associated with bills that have a close correspondence to the first dimension, since there is some variation in the degree to which the first dimension common space characterizes the legislation included in this study. Specifically, we repeat our estimates for conferences for which the first dimension common space correctly classifies at least 80% of the roll call votes on at least one key vote—on final passage or the conference report. This was an arbitrarily selected threshold that provides a high level of classification as well as a sufficient number of observations for the analysis. While this requirement severely reduces the number of observations (n = 880), we can be quite confident that the dimension most relevant to the parties is of primary consideration for these conferences. We find no substantive differences between these results and the ones to follow. These results are available from the authors upon request.

8. While the typology was originally introduced for House committees, we apply the typology to the Senate using the Senate committees with equivalent jurisdictions to the House committees. The House “prestige” category includes Appropriations, Budget, Rules, and Ways and Means, and the Senate “prestige” category includes Appropriations, Budget, and Finance. The House “policy” category includes Banking, Commerce, DC, Education, Government Reform, International Affairs, and Judiciary, and the Senate “policy” category includes Banking, Commerce, DC, Education, Government Reform, Foreign Relations, and Judiciary. The categorization for the Senate committees appears consistent with their breadth of influence and importance, as evidenced by the desirability of the committees (see Groseclose and Stewart 1999).

9. We used a number of alternative measures to tap conference delegation influence, including Groseclose and Stewart (“Grosewart”) scores (Groseclose and Stewart 1998, 1999), entropy scores, span scores (for a detailed explanation of entropy and span scores, see Sheingate 2006), and a principal component combination of these measures. We arrived at substantively similar results to those below using the alternative measures. Results are available upon request.

10. Alternatively, we also included the separate components (distance between party medians and the standard deviations of majority and minority party preferences) in lieu of our single polarization variable, and found substantively similar results.

11. Variations in this measure, including the use of a quadratic difference of medians in the numerator and the square root of the product of standard deviations in the denominator, result in substantively identical results to those following.

12. We estimated separate models including interaction terms between polarization and the party eras variables. We found that the polarization variable remains positive and statistically significant, while the interaction terms do not achieve statistical significance. This suggests that the effect of polarization is robust to the periods analyzed. We used multiple model comparison methods (Bayesian information criterion, Akaike information criterion, and likelihood ratio test), and found strong support for the models excluding the interaction terms. Moreover, we believe there is little theoretical reason to expect polarization to have different effects across periods, as would be suggested by the inclusion of the interactions. Rather, we posit that the linear relationship between polarization and pro-majority bias does not change across time, but the ability of majority parties to pursue partisan influence for any given level of polarization is related to the institutions governing conference appointments. Therefore, we believe that this story is best captured by the inclusion of the individual variables only.

13. For each chamber, we calculate the overall support percentage by Congress from the data made available by George C. Edwards III (for a detailed discussion of the measure, see Edwards 1989).

14. Conference committees that exhibit pro-minority party bias (i.e., conferences with medians situated on the minority party side of the chamber median) appear to be a function of both the party ratios in conference and the ideological composition of the parent standing committees. For conferences with pro-minority party bias, the party ratio in conference is statistically more equitable than conferences that exhibit pro-majority party bias (difference of means test is significant at the p < 0.05 level). In addition, pro-minority conferences are associated with standing committees with substantially less pro-majority bias than other conferences (difference of means test is significant at p < 0.01 level). A multivariate analysis of pro-minority conferences suggests that the composition of the parent committee is the strongest predictor of such conferences.

15. We do not include the 107th Senate (2001–2002) in Figure 2 due to the changes in majority party control that occurred during this Congress.

16. We note that the intercept for the House equation is negative and statistically significant, indicating Democratic majority party influence, when we remove the 88th and 89th Congresses (1963–1966) from the estimation. Conservative Democrats, by virtue of seniority, disproportionately dominated conferences in these Congresses. In later Democratic-controlled Congresses, Speakers took greater liberties in appointing pro-majority conference delegations.

17. We find considerable improvement in model fit across all equations when including the variable controlling for pro-majority bias in the parent committee. For the equations appearing in Table 2, we find a difference in the Bayesian information criterion of: (1) 255.70, (2) 236.51, (3) 216.85, (4) 987.65, (5) 1037.75, and (6) 1013.47. The even stronger improvement in model fit for the Senate equations (4-6) further emphasizes the close relationship between parent committee and conference delegation bias in the Senate.

18. The effect of polarization in the 107th Congress (2001–2002) is approximately .14 units in the common space, since the value of polarization is 5.41. This effect is roughly equal to the difference in common space scores between Representative Roukema (R-NJ) and Representative Hyde (R-IL), who are separated on the first dimension common space by approximately 10.3% of the membership.

19. We use Howell, Adler, Cameron, and Riemann's (2000) coding of legislative significance to assess the importance of the legislation associated with the conferences included in the analysis. We then estimate the models appearing in equations 2 and 3 of Table 2. The effect of the Seat Ratio variable falls by nearly one-half for legislation categorized in the top two significance categories (“A” and “B”) compared to the bottom two significance categories (“C” and “D”). We note that the Seat Ratio variable is virtually unchanged for the analogous Senate analysis. The results of this preliminary analysis are available from the authors upon request.

20. Senate predictions are available from the authors upon request.